Parallel retrieval and application to conceptual knowledge

Humans have the ability to recognize and reason about a wide variety of social behaviors in their fellows: deception, professional relationships, goal seeking, argumentation, etc. Much of this ability is centered around our ability to manipulate conceptual knowledge. There have been a number of computer programs which have demonstrated the capability to use conceptual knowledge as modeled by data structures in symbolic languages. Three fundamental problems are knowledge representation, knowledge access (i.e. which knowledge structures to activate) and knowledge application to specific instances (i.e. which symbols or roles in the knowledge structures are bound to which objects in the current situation). Although results have been enlightening as to processes underlying the use of conceptual knowledge, they have all been based on serial examination of symbolic data structures. This serial approach is unacceptable as a complete model of human performance. This paper presents a distributed connectionist model which retrieves and applies conceptual structures in parallel.